3

I'm studying the change of variable formula in Lebesgue integral, and I've found some similar materials in here.

Among the explanation, I'm not sure why $F \circ g$ is differentiable a.e

If it meant that $F \circ g$ is absolutely continuous, I guess there is some counter-example that makes the composition not absolutely continuous such as this

What property makes the composition differentiable a.e? Any comments about this including my error (if exists) would be grateful. Thank you.

-additional-

I have additional questions regarding the answer.

1: I suppose when $f$ is real-valued function (actually I'm interested in this general case), I could define $f_n$ as follows and then apply rest of your methods (including DCT). Tell me if this is wrong. $$f_n(x) = \begin{cases}n &\text{ if } f(x) \ge n\\ -n &\text{ if } f(x) \le -n\\ f(x) &\text{ otherwise}\end{cases} $$

2: I think the prerequisite for this question is estimated to be the following statement; $f$ is integrable, and $g$ is absolutely continuous and increasing, then $f(g(x))g'(x)$ is integrable as well. This was my trial, which seems not perfect.

Suppose $g(x) = y$, then this leads to $g'(x)dx = dy$ (basic calculus). Then,

$\int |f(g(x))g'(x)|dx$

=$\int |f(y)|dy$ < $\infty$ (∵ $f$ is integrable)

But this doesn't look nice to show the integrability of $f(g(x))g'(x)$. Is there any other way to show $f(g(x))g'(x)$ is integrable? Or, is this enough to prove the integrability?

jason 1
  • 727

1 Answers1

1

Using notations from the question you originally cite, define the functions $F : g([a,b])\to\mathbb R$ and $G:[a,b]\to\mathbb R$ by

$$F(x) = \int_{g(a)}^x f(t)\ dt\ \text{ and }\ G(x)=\int_a^x f(g(t))g'(t)\ dt $$

Where $g:[a,b]\to\mathbb R$ is a strictly increasing absolutely continuous function and $f:[c,d]\supseteq g([a,b])\to\mathbb R$ is non-negative and integrable.

Now for any integer $n\ge 1$, denote by $f_n$ the truncation of $f$ to level $n$, i.e. $$f_n(x) := f(x) \wedge n = \begin{cases}n &\text{ if } f(x)\ge n\\ f(x) &\text{ otherwise}\end{cases} $$

And accordingly denote by $F_n$ and $G_n$ the integrals $F_n(x) := \int_{g(a)}^x f_n(t)\ dt $ and $G_n(x) := \int_a^x f_n(g(t))g'(t)\ dt$.

Now for any $n\ge 1$ and $y, x\in g([a,b])$, we have

$$\begin{align*}|F_n(y)-F_n(x)|&=\left|\int_x^y f_n(t)\ dt\right|\\ &\leq\int_x^y|f_n(t)|\ dt\\ &\leq\int_x^y n\ dt\le n|y-x| \end{align*} $$

Hence $F_n$ is Lipschitz continuous and therefore $F_n\circ g$ is absolutely continuous, hence a.e. differentiable. Now you can argue as in the previous answer to see that $F_n\circ g (b) = G_n(b)$, that is $$\int_{g(a)}^{g(b)} f_n(t)\ dt = \int_a^b f_n(g(t))g'(t)\ dt $$ Now you can apply the Dominated Convergence Theorem and conclude by letting $n\to\infty$.

  • Thank you for your answer. I've got more questions about it, which is written in the main question below the "-additional-". – jason 1 May 09 '23 at 04:09
  • You're welcome. Regarding your question 1., yes that's how you would define $f_n$ if you drop the non-negativity requirement. Regarding your point 2. you can argue by Fatou's lemma : $$\int_a^b f(g(t)) g'(t)\ dt \le \lim\inf_n \int_a^b f_n(g(t))g'(t) dt = \int_{g(a)}^{g(b)} f(t) dt <\infty $$ – Stratos supports the strike May 09 '23 at 17:46
  • Thanks, but after I read the whole explanations, I found your comment is based on your final result $\int f_n(t) dt = \int f_n(g(t))g'(t) dt$. Before this final result, I think the integrability of $f(g(t))g'(t)$ should be proved first. But base on my understanding, $\int f_n(t) dt = \int f_n(g(t))g'(t) dt$, as well as the expression in your last comment, already uses the fact that $f(g(t))g'(t)$ is integrable. Is there any other way to show the integrability of $f(g(t))g'(t)$ first? – jason 1 May 13 '23 at 06:49
  • And, if you've got some time to add some comments, you may use this link; https://math.stackexchange.com/q/4698200/1118138 . Thank you. – jason 1 May 13 '23 at 07:16
  • The identity $\int f_n(t) dt = \int f_n(g(t))g'(t) dt$ did not rely on the integrability of $f(g(t))g'(t)$, so there is no need to "prove the integrability of $f(g(t))g'(t)$ first" to deduce it. Once that identity is shown, Fatou's lemma as applied in my above comment shows that $f(g(t))g'(t)$ is indeed integrable (and DCT shows that its integral is actually equal to that of $f(t)$)... I see that you have opened another question about the integrability issue, so if you have further doubts about what I just said just raise them there, as this is beyond the scope of your original question here. – Stratos supports the strike May 13 '23 at 15:12
  • (Side note : please do not edit your question to add "additional questions" after your original question has been answered. Just create new questions instead. See also this, this and this on meta) – Stratos supports the strike May 13 '23 at 15:15
  • Ah, alright, now I can see. Thank you for your answer. – jason 1 May 14 '23 at 05:56